[None][infra] Add a probe workflow for running Check Test List on Git…

.github/workflows/check-test-list-probe.yml

@@ -0,0 +1,141 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Probe how long the Jenkins "Check Test List" stage takes on a GitHub-hosted
+# runner with CUDA torch, the full requirements-dev set and a pip cache.
+name: Check Test List Probe
+on:
+  pull_request:
+    paths:
+      - '.github/workflows/check-test-list-probe.yml'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
+  cancel-in-progress: true
+
+env:
+  TORCH_VERSION: '2.14.0'
+  TORCH_INDEX: https://download.pytorch.org/whl/cu132
+
+jobs:
+  collect:
+    name: Collect (full-cuda deps, ubuntu-latest)
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    steps:
+      # Same layout as the Jenkins stage: the source lives in a subdirectory and
+      # check_test_list.py is invoked by absolute path from the workspace root.
+      - uses: actions/checkout@v6
+        with:
+          path: llm
+
+      - uses: actions/setup-python@v6
+        id: python
+        with:
+          python-version: '3.12'
+          cache: 'pip'
+          cache-dependency-path: |
+            llm/requirements.txt
+            llm/requirements-dev.txt
+            llm/constraints.txt
+
+      - name: Install torch
+        run: |
+          start=$SECONDS
+          python -m pip install --upgrade pip
+          python -m pip install --index-url "${TORCH_INDEX}" "torch==${TORCH_VERSION}" torchvision
+          echo "TORCH_SECONDS=$((SECONDS - start))" >> "$GITHUB_ENV"
+
+      # mpi4py loads libmpi when mpi4py.MPI is imported. The CI image ships
+      # HPC-X OpenMPI; on a plain runner the openmpi wheel provides it.
+      - name: Install requirements
+        working-directory: llm
+        run: |
+          printf 'torch==%s\n' "${TORCH_VERSION}" > "${RUNNER_TEMP}/torch-pin.txt"
+          start=$SECONDS
+          python -m pip install --extra-index-url "${TORCH_INDEX}" \
+            -c "${RUNNER_TEMP}/torch-pin.txt" -r requirements-dev.txt openmpi
+          echo "REQ_SECONDS=$((SECONDS - start))" >> "$GITHUB_ENV"
+
+      # urm.nvidia.com does not resolve from GitHub-hosted runners. check_test_list
+      # only runs `trt-test-db -d DIR --test-names --output FILE`, which lists every
+      # test name in DIR/*.yml; this stand-in produces the same set from the YAML.
+      - name: Install a trt-test-db stand-in
+        run: |
+          mkdir -p "${RUNNER_TEMP}/bin"
+          cat > "${RUNNER_TEMP}/bin/trt-test-db" <<'EOF'
+          #!/usr/bin/env python3
+          import argparse
+          import glob
+          import os
+
+          import yaml
+
+          parser = argparse.ArgumentParser()
+          parser.add_argument("-d", required=True)
+          parser.add_argument("--test-names", action="store_true", required=True)
+          parser.add_argument("--output", required=True)
+          args = parser.parse_args()
+          names = set()
+          for path in glob.glob(os.path.join(args.d, "*.yml")):
+              with open(path) as f:
+                  for key, blocks in (yaml.safe_load(f) or {}).items():
+                      if isinstance(blocks, list):
+                          for block in blocks:
+                              names.update(str(t) for t in block.get("tests") or [])
+          with open(args.output, "w") as f:
+              f.write("\n".join(sorted(names)) + "\n")
+          EOF
+          chmod +x "${RUNNER_TEMP}/bin/trt-test-db"
+          echo "${RUNNER_TEMP}/bin" >> "$GITHUB_PATH"
+
+      - name: Run check_test_list
+        env:
+          NVIDIA_TRITON_SERVER_VERSION: '26.09'
+        run: |
+          export LLM_ROOT="$PWD/llm" LLM_BACKEND_ROOT="$PWD/llm/triton_backend"
+          start=$SECONDS
+          # Dependencies were installed by the steps above; skip the script's own
+          # pip install so the measured set is exactly the one under test.
+          python - <<'EOF'
+          import importlib.util
+          import os
+          import sys
+
+          path = os.path.join(os.environ["LLM_ROOT"], "scripts", "check_test_list.py")
+          spec = importlib.util.spec_from_file_location("check_test_list", path)
+          module = importlib.util.module_from_spec(spec)
+          spec.loader.exec_module(module)
+          module.install_python_dependencies = lambda llm_src: None
+          sys.argv = [path, "--l0", "--qa", "--waive", "--validate", "--parity",
+                      "--check-duplicate-waives"]
+          module.main()
+          EOF
+          echo "COLLECT_SECONDS=$((SECONDS - start))" >> "$GITHUB_ENV"
+
+      - name: Summary
+        if: always()
+        run: |
+          {
+            echo "## Collect with full-cuda deps on ubuntu-latest"
+            echo "pip cache restored: ${{ steps.python.outputs.cache-hit }}; size now: $(du -sh "$(python -m pip cache dir)" | cut -f1)"
+            echo
+            echo "| Step | Seconds |"
+            echo "|---|---|"
+            echo "| torch | ${TORCH_SECONDS:-} |"
+            echo "| requirements | ${REQ_SECONDS:-} |"
+            echo "| check_test_list | ${COLLECT_SECONDS:-} |"
+          } | tee -a "$GITHUB_STEP_SUMMARY"